Claude Prompt to Analyze a CSV Dataset and Summarize Key Insights
This Claude prompt is for analysts, founders, and marketers who need to turn a raw CSV export into a plain-language summary of what the data actually shows. It's built for situations where you have a spreadsheet of sales, survey, or usage data and need the key trends, outliers, and takeaways without writing your own analysis script first.
The prompt asks the model to work only from the data you paste in or describe, state its assumptions about column meanings, and separate factual observations from suggested next steps. That distinction matters because a language model does not run calculations internally the way a spreadsheet formula does — it reasons over the numbers you give it, so accuracy depends on giving it clean, complete data and asking it to show its reasoning.
Because it's written as plain role and task instructions, this prompt works in Claude, ChatGPT, or Gemini with no changes. For very large datasets, pair it with each tool's file upload or code-execution feature rather than pasting raw rows into the chat.
Prompt template
Role: You are a data analyst who explains findings in plain language for a non-technical audience. Context: - Dataset description: [WHAT THE DATA REPRESENTS, E.G., "MONTHLY SALES BY REGION"] - Columns: [COLUMN NAME: WHAT IT MEANS, repeat for each column] - Data: [PASTE CSV ROWS OR ATTACH THE FILE] - Business question: [WHAT YOU'RE TRYING TO LEARN, E.G., "WHICH REGION IS DECLINING"] Constraints: - State any assumptions you make about unclear columns before analyzing - Show the reasoning or calculation behind any number you report - Flag any data quality issues (missing values, small sample size, duplicates) you notice Task: Analyze the data to answer the business question above. Output format: 1. Key findings (3-5 bullet points, plain language) 2. Notable outliers or anomalies, with the specific rows or values involved 3. Caveats or data quality issues 4. Suggested next steps for further analysis
Example input
Dataset description: Monthly subscription sales by region for Q1 2026 Columns: month: calendar month; region: sales region (North, South, East, West); new_subscribers: count of new paid subscriptions; churned_subscribers: count of cancellations Data: month,region,new_subscribers,churned_subscribers Jan,North,120,30 Jan,South,95,25 Jan,East,110,28 Jan,West,60,22 Feb,North,118,32 Feb,South,90,27 Feb,East,105,26 Feb,West,40,24 Mar,North,125,29 Mar,South,88,30 Mar,East,100,25 Mar,West,22,26 Business question: Which region is at risk of net subscriber decline?
Example output
1. Key findings: - The West region shows a clear downward trend in new subscribers, dropping from 60 in January to 22 in March, while churn stayed roughly flat around 22-26. - North, South, and East kept new subscribers consistently above churned subscribers each month. - West is the only region where net new subscribers turned negative in March (22 new vs. 26 churned). 2. Notable outliers: - West, March: new_subscribers (22) fell below churned_subscribers (26), a net loss of 4 subscribers, the only negative month in the dataset. 3. Caveats: - Only three months of data are included, which isn't enough to confirm a long-term trend. - The dataset doesn't include reasons for churn or new-subscriber acquisition channel, so the cause of West's decline can't be determined from this data alone. 4. Suggested next steps: - Pull West's acquisition channel data to see if a specific campaign or channel dropped off. - Extend the date range to confirm whether this is a trend or a one-month dip. - Compare West's churn reasons against other regions to check for a region-specific cause.
When to use it
- Getting a first-pass summary of a sales, survey, or product-usage export before a meeting
- Spotting outliers or unexpected patterns in a dataset you haven't explored yet
- Turning a raw CSV into a short written summary for a non-technical stakeholder
- Generating hypotheses to test further in a spreadsheet or BI tool, not as a final report
Best practices
- Include column headers and a short description of what each column represents
- Ask the model to state any assumptions it made about ambiguous columns before analyzing
- Request that it separate observed facts from suggested interpretations or next steps
- For datasets over a few hundred rows, use the tool's file upload or code execution feature instead of pasting raw text
Common mistakes
- Pasting a partial or truncated CSV and expecting totals or averages to be accurate
- Not specifying the business question, which leads to a generic, unfocused summary
- Treating the output as verified statistics instead of asking for the calculation shown
- Skipping a request for caveats, so small sample sizes or missing data go unmentioned
FAQs
Can Claude or ChatGPT actually calculate statistics from a CSV?
They reason over the numbers you provide and can perform straightforward arithmetic, but for large datasets or precise statistics, use the tool's code execution or file analysis feature rather than relying on the chat response alone.
How much data can I paste directly into the prompt?
For more than a few hundred rows, accuracy drops and you risk hitting context limits. Upload the file directly if the tool supports it, or ask the model to analyze a representative sample.
Why does the prompt ask the model to state its assumptions?
Column names are often ambiguous (like "active" or "score"), and having the model state its interpretation upfront lets you catch a misread before it affects the whole analysis.
Does this prompt work the same in Claude, ChatGPT, and Gemini?
Yes, it uses plain instructions with no vendor-specific syntax, though each tool's file upload and code execution capabilities differ, so check what's available before pasting large datasets.